The challenge
A food publishing brand had built a successful recipe platform on a tightly coupled CMS. It worked — until it didn't. As ambitions grew to include AI-assisted publishing, multi-domain content, and premium membership features, the existing architecture became the obstacle.
The CMS layer had spread throughout the codebase: website rendering, content querying, image pipelines, admin tooling, and AI-assisted editing all had direct dependencies on it. Migrating, extending, or replacing any part of the system required touching everything else.
The business risks were real: difficult migration paths, vendor lock-in, fragile editorial tooling, and a hard ceiling on how far the platform could grow. The client wanted to evolve from a recipe website into a full creator-content platform — and the existing architecture couldn't take them there.
Our solution
We designed and implemented a modular, API-first content platform built around a custom headless CMS foundation. The guiding principle was clean separation: each layer of the system does one job, communicates through well-defined interfaces, and can evolve independently.
The architecture breaks down into five distinct layers:
- Frontend website layer — Next.js, statically generated, SEO-optimised
- SDK abstraction layer — decouples the frontend from any specific CMS implementation
- API and content platform — handles all content CRUD, publishing workflows, and relations
- AI orchestration layer — sits alongside editorial tooling, not bolted on after the fact
- Asset management pipeline — portable image handling with CDN-ready delivery
AI built into publishing, not added on top
The AI layer was integrated directly into the editorial experience from day one — not treated as a future add-on. Editors can access AI assistance within the same interface they use to write and publish, which drives actual adoption rather than tool-switching.
AI Editorial Assistance
Recipe editing, category suggestions, content refinement, and structured publishing validation — all AI-assisted from within the editor.
Structured Recipe Engine
Rich domain model covering ingredient groups, ordered steps, nutrition metadata, FAQs, SEO fields, and related recipes — with atomic publishing for consistency.
Portable Media Pipeline
Image upload, metadata extraction, asset deduplication, and CDN-ready delivery — all through a portable abstraction that replaced the CMS-specific image tooling.
SEO-First Frontend
Statically generated pages with structured metadata, Open Graph optimisation, and search-friendly URL structures — built into the architecture, not retrofitted.
Multi-domain from the start
One of the key design decisions was to build a content platform rather than a recipe website. The data model was engineered to support multiple content verticals — recipes, gardening activities, fitness content — without requiring schema rewrites each time a new domain is added. All three verticals were live on the same infrastructure from launch.
Subscription and entitlement infrastructure
We implemented enterprise-grade feature and access controls from the outset: feature flag management, usage limits, premium feature gating, AI usage quotas, image storage quotas, and group-based permissions. This positions the client to launch tiered memberships and monetise premium content without a significant re-architecture.
Business outcomes
- Eliminated vendor lock-in — the platform is portable and no longer dependent on any single CMS provider
- Faster feature delivery — clean layer separation means new features touch one layer, not the entire stack
- AI-ready infrastructure — the orchestration layer is built for expansion as AI capabilities mature
- Multi-domain growth — recipes, gardening, and fitness content all run on shared infrastructure
- Premium product foundation — entitlement infrastructure is in place and ready to activate
- Reusable platform IP — the client owns a publishing platform, not just a website